手肘超音波圖像增強方法之比較

Abstract

圖片的品質是醫學相關應用中超音波影像的關鍵組成部分，但它可能會受到人為操作和可用圖片數量限制的影響。我們的研究旨在比較相對較少圖片的手肘肱骨超音波影像在12種不同的圖像增強方法中，在圖片閾值處理、圖片過濾和圖片形態學三個組別下的圖片品質。我們使用了三種常用的圖片品質評估指標，包括峰值信噪比（PSNR）、結構相似性指數（SSIM）和信噪比（SNR）來評估這些方法。 在這12種圖像增強方法中，圖片過濾組中的雙邊濾波器（BILATERALFILTER）和高斯模糊（GAUSSIANBLUR）在PSNR、SSIM和SNR這三個指標上優於其他方法。無論是3、5還是9的內核大小，這兩種方法在性能上始終表現良好。這表明我們的研究成功地找到了基於這些指標確保圖片品質的方法。此外，雙邊濾波器和高斯模糊方法在數據增強和深度學習方法中也顯示出潛力，可以解決特定器官和組織的超音波影像數量不足的問題。

Image quality is a critical component of ultrasound images for medical related application, but it can be affected by human operation and the limited number of available images. Our research aims to compare the quality of ultrasound images of humerus in the elbow, which has relatively few images, between 12 different image enhancement methods under three groups of image thresholding, image filtering and image morphology. We evaluate these methods using three commonly used three metrics for image quality such as PSNR (Peak Signal-to-Noise Ratio), SSIM (Structural Similarity Index), and SNR (Signal-to-Noise Ratio). Among the 12 image enhancement methods, the BILATERALFILTER and GAUSSIANBLUR under the image filtering group outperform the other methods across all three metrics of PSNR, SSIM, and SNR. Regardless of the kernel size of 3, 5, or 9, the two methods also consistently exhibit better performance than other methods. This demonstrates that our research has successfully identified methods that ensure image quality based on these metrics. Furthermore, the BILATERALFILTER and GAUSSIANBLUR methods show potential for use in data augmentation for deep learning approaches to resolve the insufficiently available ultrasound images for specific organs and tissues.